Clustering Research Based On Support Vector Machine

With the advancement of science and Internet technology,many aspects of people's lives have established connections with the Internet,and more and more Internet applications and Internet products have entered people's lives.These changes have made everyone's life more efficient and faster while generating plenty of network data.The valuable knowledge and information hidden in big data can not be ignored for business and scientific research in various fields.Therefore,the mining and learning of big data has gradually become a hot spot in the current era.How to reasonably and effectively mine the potential information in data has become the focus of more and more people.Clustering and classification are two significant tasks in the machine learning field,both of them have many proven techniques and successful real-world applications.Clustering aggregates similar data to form several clusters based on certain similarity criterion,whereas classification determines the class of unknown data based on classification rules learned from data with clear class features.Inspired by the ideas of clustering and classification,this thesis presents a new clustering technique,called seeds-based support vector machine clustering(SSVMC),which successfully integrates classification elements into clustering.SSVMC first uses a clustering method to obtain an initial clustering result,takes data points with strong class features in each cluster as seed points,learns classification rules from these seed points while using the idea of clustering to find cluster centers,and finally uses the learned rules to classify all the data to obtain clusters.Experiments conducted on 16 standard datasets demonstrate the effectiveness of the proposed unsupervised SSVMC over certain state-of-the-art unsupervised algorithms.